2 * Definitions for the 'struct sk_buff' memory handlers.
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
17 #include <linux/kernel.h>
18 #include <linux/compiler.h>
19 #include <linux/time.h>
20 #include <linux/bug.h>
21 #include <linux/cache.h>
22 #include <linux/rbtree.h>
23 #include <linux/socket.h>
24 #include <linux/refcount.h>
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
44 /* The interface for checksum offload between the stack and networking drivers
47 * A. IP checksum related features
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
54 * The checksum related features are:
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
92 * Device did not checksum this packet e.g. due to lack of capabilities.
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 * CHECKSUM_UNNECESSARY:
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
112 * FCOE: indicates the CRC in FC frame has been validated.
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
152 * The driver is required to checksum the packet as seen by hard_start_xmit()
153 * from skb->csum_start up to the end, and to record/write the checksum at
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
176 * The skb was already checksummed by the protocol, or a checksum is not
179 * CHECKSUM_UNNECESSARY:
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
188 * D. Non-IP checksum (CRC) offloads
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
208 * E. Checksumming on output with GSO.
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE 0
221 #define CHECKSUM_UNNECESSARY 1
222 #define CHECKSUM_COMPLETE 2
223 #define CHECKSUM_PARTIAL 3
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL 3
228 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X) \
230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
243 struct pipe_inode_info;
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
257 BRNF_PROTO_UNCHANGED,
265 struct net_device *physindev;
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
270 /* prerouting: detect dnat in orig/reply direction */
272 struct in6_addr ipv6_daddr;
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
278 char neigh_header[8];
283 struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
294 /* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
298 * Since GRO uses frags we allocate at least 16 regardless of page
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
306 extern int sysctl_max_skb_frags;
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
311 #define GSO_BY_FRAGS 0xFFFF
313 typedef struct skb_frag_struct skb_frag_t;
315 struct skb_frag_struct {
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
348 static inline bool skb_frag_must_loop(struct page *p)
350 #if defined(CONFIG_HIGHMEM)
358 * skb_frag_foreach_page - loop over pages in a fragment
360 * @f: skb frag to operate on
361 * @f_off: offset from start of f->page.p
362 * @f_len: length from f_off to loop over
363 * @p: (temp var) current page
364 * @p_off: (temp var) offset from start of current page,
365 * non-zero only on first page.
366 * @p_len: (temp var) length in current page,
367 * < PAGE_SIZE only on first and last page.
368 * @copied: (temp var) length so far, excluding current p_len.
370 * A fragment can hold a compound page, in which case per-page
371 * operations, notably kmap_atomic, must be called for each
374 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
375 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
376 p_off = (f_off) & (PAGE_SIZE - 1), \
377 p_len = skb_frag_must_loop(p) ? \
378 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
381 copied += p_len, p++, p_off = 0, \
382 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
384 #define HAVE_HW_TIME_STAMP
387 * struct skb_shared_hwtstamps - hardware time stamps
388 * @hwtstamp: hardware time stamp transformed into duration
389 * since arbitrary point in time
391 * Software time stamps generated by ktime_get_real() are stored in
394 * hwtstamps can only be compared against other hwtstamps from
397 * This structure is attached to packets as part of the
398 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
400 struct skb_shared_hwtstamps {
404 /* Definitions for tx_flags in struct skb_shared_info */
406 /* generate hardware time stamp */
407 SKBTX_HW_TSTAMP = 1 << 0,
409 /* generate software time stamp when queueing packet to NIC */
410 SKBTX_SW_TSTAMP = 1 << 1,
412 /* device driver is going to provide hardware time stamp */
413 SKBTX_IN_PROGRESS = 1 << 2,
415 /* device driver supports TX zero-copy buffers */
416 SKBTX_DEV_ZEROCOPY = 1 << 3,
418 /* generate wifi status information (where possible) */
419 SKBTX_WIFI_STATUS = 1 << 4,
421 /* This indicates at least one fragment might be overwritten
422 * (as in vmsplice(), sendfile() ...)
423 * If we need to compute a TX checksum, we'll need to copy
424 * all frags to avoid possible bad checksum
426 SKBTX_SHARED_FRAG = 1 << 5,
428 /* generate software time stamp when entering packet scheduling */
429 SKBTX_SCHED_TSTAMP = 1 << 6,
432 #define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
433 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
435 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
438 * The callback notifies userspace to release buffers when skb DMA is done in
439 * lower device, the skb last reference should be 0 when calling this.
440 * The zerocopy_success argument is true if zero copy transmit occurred,
441 * false on data copy or out of memory error caused by data copy attempt.
442 * The ctx field is used to track device context.
443 * The desc field is used to track userspace buffer index.
446 void (*callback)(struct ubuf_info *, bool zerocopy_success);
462 struct user_struct *user;
467 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
469 int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
470 void mm_unaccount_pinned_pages(struct mmpin *mmp);
472 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
473 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
474 struct ubuf_info *uarg);
476 static inline void sock_zerocopy_get(struct ubuf_info *uarg)
478 refcount_inc(&uarg->refcnt);
481 void sock_zerocopy_put(struct ubuf_info *uarg);
482 void sock_zerocopy_put_abort(struct ubuf_info *uarg);
484 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
486 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
487 struct msghdr *msg, int len,
488 struct ubuf_info *uarg);
490 /* This data is invariant across clones and lives at
491 * the end of the header data, ie. at skb->end.
493 struct skb_shared_info {
498 unsigned short gso_size;
499 /* Warning: this field is not always filled in (UFO)! */
500 unsigned short gso_segs;
501 struct sk_buff *frag_list;
502 struct skb_shared_hwtstamps hwtstamps;
503 unsigned int gso_type;
507 * Warning : all fields before dataref are cleared in __alloc_skb()
511 /* Intermediate layers must ensure that destructor_arg
512 * remains valid until skb destructor */
513 void * destructor_arg;
515 /* must be last field, see pskb_expand_head() */
516 skb_frag_t frags[MAX_SKB_FRAGS];
519 /* We divide dataref into two halves. The higher 16 bits hold references
520 * to the payload part of skb->data. The lower 16 bits hold references to
521 * the entire skb->data. A clone of a headerless skb holds the length of
522 * the header in skb->hdr_len.
524 * All users must obey the rule that the skb->data reference count must be
525 * greater than or equal to the payload reference count.
527 * Holding a reference to the payload part means that the user does not
528 * care about modifications to the header part of skb->data.
530 #define SKB_DATAREF_SHIFT 16
531 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
535 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
536 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
537 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
541 SKB_GSO_TCPV4 = 1 << 0,
543 /* This indicates the skb is from an untrusted source. */
544 SKB_GSO_DODGY = 1 << 1,
546 /* This indicates the tcp segment has CWR set. */
547 SKB_GSO_TCP_ECN = 1 << 2,
549 SKB_GSO_TCP_FIXEDID = 1 << 3,
551 SKB_GSO_TCPV6 = 1 << 4,
553 SKB_GSO_FCOE = 1 << 5,
555 SKB_GSO_GRE = 1 << 6,
557 SKB_GSO_GRE_CSUM = 1 << 7,
559 SKB_GSO_IPXIP4 = 1 << 8,
561 SKB_GSO_IPXIP6 = 1 << 9,
563 SKB_GSO_UDP_TUNNEL = 1 << 10,
565 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
567 SKB_GSO_PARTIAL = 1 << 12,
569 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
571 SKB_GSO_SCTP = 1 << 14,
573 SKB_GSO_ESP = 1 << 15,
575 SKB_GSO_UDP = 1 << 16,
578 #if BITS_PER_LONG > 32
579 #define NET_SKBUFF_DATA_USES_OFFSET 1
582 #ifdef NET_SKBUFF_DATA_USES_OFFSET
583 typedef unsigned int sk_buff_data_t;
585 typedef unsigned char *sk_buff_data_t;
589 * struct sk_buff - socket buffer
590 * @next: Next buffer in list
591 * @prev: Previous buffer in list
592 * @tstamp: Time we arrived/left
593 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
594 * @sk: Socket we are owned by
595 * @dev: Device we arrived on/are leaving by
596 * @cb: Control buffer. Free for use by every layer. Put private vars here
597 * @_skb_refdst: destination entry (with norefcount bit)
598 * @sp: the security path, used for xfrm
599 * @len: Length of actual data
600 * @data_len: Data length
601 * @mac_len: Length of link layer header
602 * @hdr_len: writable header length of cloned skb
603 * @csum: Checksum (must include start/offset pair)
604 * @csum_start: Offset from skb->head where checksumming should start
605 * @csum_offset: Offset from csum_start where checksum should be stored
606 * @priority: Packet queueing priority
607 * @ignore_df: allow local fragmentation
608 * @cloned: Head may be cloned (check refcnt to be sure)
609 * @ip_summed: Driver fed us an IP checksum
610 * @nohdr: Payload reference only, must not modify header
611 * @pkt_type: Packet class
612 * @fclone: skbuff clone status
613 * @ipvs_property: skbuff is owned by ipvs
614 * @tc_skip_classify: do not classify packet. set by IFB device
615 * @tc_at_ingress: used within tc_classify to distinguish in/egress
616 * @tc_redirected: packet was redirected by a tc action
617 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
618 * @peeked: this packet has been seen already, so stats have been
619 * done for it, don't do them again
620 * @nf_trace: netfilter packet trace flag
621 * @protocol: Packet protocol from driver
622 * @destructor: Destruct function
623 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
624 * @_nfct: Associated connection, if any (with nfctinfo bits)
625 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
626 * @skb_iif: ifindex of device we arrived on
627 * @tc_index: Traffic control index
628 * @hash: the packet hash
629 * @queue_mapping: Queue mapping for multiqueue devices
630 * @xmit_more: More SKBs are pending for this queue
631 * @ndisc_nodetype: router type (from link layer)
632 * @ooo_okay: allow the mapping of a socket to a queue to be changed
633 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
635 * @sw_hash: indicates hash was computed in software stack
636 * @wifi_acked_valid: wifi_acked was set
637 * @wifi_acked: whether frame was acked on wifi or not
638 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
639 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
640 * @dst_pending_confirm: need to confirm neighbour
641 * @napi_id: id of the NAPI struct this skb came from
642 * @secmark: security marking
643 * @mark: Generic packet mark
644 * @vlan_proto: vlan encapsulation protocol
645 * @vlan_tci: vlan tag control information
646 * @inner_protocol: Protocol (encapsulation)
647 * @inner_transport_header: Inner transport layer header (encapsulation)
648 * @inner_network_header: Network layer header (encapsulation)
649 * @inner_mac_header: Link layer header (encapsulation)
650 * @transport_header: Transport layer header
651 * @network_header: Network layer header
652 * @mac_header: Link layer header
653 * @tail: Tail pointer
655 * @head: Head of buffer
656 * @data: Data head pointer
657 * @truesize: Buffer size
658 * @users: User count - see {datagram,tcp}.c
664 /* These two members must be first. */
665 struct sk_buff *next;
666 struct sk_buff *prev;
669 struct net_device *dev;
670 /* Some protocols might use this space to store information,
671 * while device pointer would be NULL.
672 * UDP receive path is one user.
674 unsigned long dev_scratch;
677 struct rb_node rbnode; /* used in netem & tcp stack */
686 * This is the control buffer. It is free to use for every
687 * layer. Please put your private variables there. If you
688 * want to keep them across layers you have to do a skb_clone()
689 * first. This is owned by whoever has the skb queued ATM.
691 char cb[48] __aligned(8);
695 unsigned long _skb_refdst;
696 void (*destructor)(struct sk_buff *skb);
698 struct list_head tcp_tsorted_anchor;
704 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
707 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
708 struct nf_bridge_info *nf_bridge;
715 /* Following fields are _not_ copied in __copy_skb_header()
716 * Note that queue_mapping is here mostly to fill a hole.
720 /* if you move cloned around you also must adapt those constants */
721 #ifdef __BIG_ENDIAN_BITFIELD
722 #define CLONED_MASK (1 << 7)
724 #define CLONED_MASK 1
726 #define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
728 __u8 __cloned_offset[0];
735 __unused:1; /* one bit hole */
737 /* fields enclosed in headers_start/headers_end are copied
738 * using a single memcpy() in __copy_skb_header()
741 __u32 headers_start[0];
744 /* if you move pkt_type around you also must adapt those constants */
745 #ifdef __BIG_ENDIAN_BITFIELD
746 #define PKT_TYPE_MAX (7 << 5)
748 #define PKT_TYPE_MAX 7
750 #define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
752 __u8 __pkt_type_offset[0];
762 __u8 wifi_acked_valid:1;
766 /* Indicates the inner headers are valid in the skbuff. */
767 __u8 encapsulation:1;
768 __u8 encap_hdr_csum:1;
770 __u8 csum_complete_sw:1;
772 __u8 csum_not_inet:1;
774 __u8 dst_pending_confirm:1;
775 #ifdef CONFIG_IPV6_NDISC_NODETYPE
776 __u8 ndisc_nodetype:2;
778 __u8 ipvs_property:1;
779 __u8 inner_protocol_type:1;
780 __u8 remcsum_offload:1;
781 #ifdef CONFIG_NET_SWITCHDEV
782 __u8 offload_fwd_mark:1;
783 __u8 offload_mr_fwd_mark:1;
785 #ifdef CONFIG_NET_CLS_ACT
786 __u8 tc_skip_classify:1;
787 __u8 tc_at_ingress:1;
788 __u8 tc_redirected:1;
789 __u8 tc_from_ingress:1;
792 #ifdef CONFIG_NET_SCHED
793 __u16 tc_index; /* traffic control index */
808 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
810 unsigned int napi_id;
811 unsigned int sender_cpu;
814 #ifdef CONFIG_NETWORK_SECMARK
820 __u32 reserved_tailroom;
824 __be16 inner_protocol;
828 __u16 inner_transport_header;
829 __u16 inner_network_header;
830 __u16 inner_mac_header;
833 __u16 transport_header;
834 __u16 network_header;
838 __u32 headers_end[0];
841 /* These elements must be at the end, see alloc_skb() for details. */
846 unsigned int truesize;
852 * Handling routines are only of interest to the kernel
854 #include <linux/slab.h>
857 #define SKB_ALLOC_FCLONE 0x01
858 #define SKB_ALLOC_RX 0x02
859 #define SKB_ALLOC_NAPI 0x04
861 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
862 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
864 return unlikely(skb->pfmemalloc);
868 * skb might have a dst pointer attached, refcounted or not.
869 * _skb_refdst low order bit is set if refcount was _not_ taken
871 #define SKB_DST_NOREF 1UL
872 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
874 #define SKB_NFCT_PTRMASK ~(7UL)
876 * skb_dst - returns skb dst_entry
879 * Returns skb dst_entry, regardless of reference taken or not.
881 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
883 /* If refdst was not refcounted, check we still are in a
884 * rcu_read_lock section
886 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
887 !rcu_read_lock_held() &&
888 !rcu_read_lock_bh_held());
889 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
893 * skb_dst_set - sets skb dst
897 * Sets skb dst, assuming a reference was taken on dst and should
898 * be released by skb_dst_drop()
900 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
902 skb->_skb_refdst = (unsigned long)dst;
906 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
910 * Sets skb dst, assuming a reference was not taken on dst.
911 * If dst entry is cached, we do not take reference and dst_release
912 * will be avoided by refdst_drop. If dst entry is not cached, we take
913 * reference, so that last dst_release can destroy the dst immediately.
915 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
917 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
918 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
922 * skb_dst_is_noref - Test if skb dst isn't refcounted
925 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
927 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
930 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
932 return (struct rtable *)skb_dst(skb);
935 /* For mangling skb->pkt_type from user space side from applications
936 * such as nft, tc, etc, we only allow a conservative subset of
937 * possible pkt_types to be set.
939 static inline bool skb_pkt_type_ok(u32 ptype)
941 return ptype <= PACKET_OTHERHOST;
944 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
946 #ifdef CONFIG_NET_RX_BUSY_POLL
953 /* decrement the reference count and return true if we can free the skb */
954 static inline bool skb_unref(struct sk_buff *skb)
958 if (likely(refcount_read(&skb->users) == 1))
960 else if (likely(!refcount_dec_and_test(&skb->users)))
966 void skb_release_head_state(struct sk_buff *skb);
967 void kfree_skb(struct sk_buff *skb);
968 void kfree_skb_list(struct sk_buff *segs);
969 void skb_tx_error(struct sk_buff *skb);
970 void consume_skb(struct sk_buff *skb);
971 void __consume_stateless_skb(struct sk_buff *skb);
972 void __kfree_skb(struct sk_buff *skb);
973 extern struct kmem_cache *skbuff_head_cache;
975 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
976 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
977 bool *fragstolen, int *delta_truesize);
979 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
981 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
982 struct sk_buff *build_skb(void *data, unsigned int frag_size);
983 static inline struct sk_buff *alloc_skb(unsigned int size,
986 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
989 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
990 unsigned long data_len,
995 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
996 struct sk_buff_fclones {
1001 refcount_t fclone_ref;
1005 * skb_fclone_busy - check if fclone is busy
1009 * Returns true if skb is a fast clone, and its clone is not freed.
1010 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1011 * so we also check that this didnt happen.
1013 static inline bool skb_fclone_busy(const struct sock *sk,
1014 const struct sk_buff *skb)
1016 const struct sk_buff_fclones *fclones;
1018 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1020 return skb->fclone == SKB_FCLONE_ORIG &&
1021 refcount_read(&fclones->fclone_ref) > 1 &&
1022 fclones->skb2.sk == sk;
1025 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1028 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1031 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1032 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1033 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1034 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1035 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1036 gfp_t gfp_mask, bool fclone);
1037 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1040 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1043 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1044 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1045 unsigned int headroom);
1046 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1047 int newtailroom, gfp_t priority);
1048 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1049 int offset, int len);
1050 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1051 int offset, int len);
1052 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1053 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1056 * skb_pad - zero pad the tail of an skb
1057 * @skb: buffer to pad
1058 * @pad: space to pad
1060 * Ensure that a buffer is followed by a padding area that is zero
1061 * filled. Used by network drivers which may DMA or transfer data
1062 * beyond the buffer end onto the wire.
1064 * May return error in out of memory cases. The skb is freed on error.
1066 static inline int skb_pad(struct sk_buff *skb, int pad)
1068 return __skb_pad(skb, pad, true);
1070 #define dev_kfree_skb(a) consume_skb(a)
1072 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1073 int getfrag(void *from, char *to, int offset,
1074 int len, int odd, struct sk_buff *skb),
1075 void *from, int length);
1077 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1078 int offset, size_t size);
1080 struct skb_seq_state {
1084 __u32 stepped_offset;
1085 struct sk_buff *root_skb;
1086 struct sk_buff *cur_skb;
1090 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1091 unsigned int to, struct skb_seq_state *st);
1092 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1093 struct skb_seq_state *st);
1094 void skb_abort_seq_read(struct skb_seq_state *st);
1096 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1097 unsigned int to, struct ts_config *config);
1100 * Packet hash types specify the type of hash in skb_set_hash.
1102 * Hash types refer to the protocol layer addresses which are used to
1103 * construct a packet's hash. The hashes are used to differentiate or identify
1104 * flows of the protocol layer for the hash type. Hash types are either
1105 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1107 * Properties of hashes:
1109 * 1) Two packets in different flows have different hash values
1110 * 2) Two packets in the same flow should have the same hash value
1112 * A hash at a higher layer is considered to be more specific. A driver should
1113 * set the most specific hash possible.
1115 * A driver cannot indicate a more specific hash than the layer at which a hash
1116 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1118 * A driver may indicate a hash level which is less specific than the
1119 * actual layer the hash was computed on. For instance, a hash computed
1120 * at L4 may be considered an L3 hash. This should only be done if the
1121 * driver can't unambiguously determine that the HW computed the hash at
1122 * the higher layer. Note that the "should" in the second property above
1125 enum pkt_hash_types {
1126 PKT_HASH_TYPE_NONE, /* Undefined type */
1127 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1128 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1129 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1132 static inline void skb_clear_hash(struct sk_buff *skb)
1139 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1142 skb_clear_hash(skb);
1146 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1148 skb->l4_hash = is_l4;
1149 skb->sw_hash = is_sw;
1154 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1156 /* Used by drivers to set hash from HW */
1157 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1161 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1163 __skb_set_hash(skb, hash, true, is_l4);
1166 void __skb_get_hash(struct sk_buff *skb);
1167 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1168 u32 skb_get_poff(const struct sk_buff *skb);
1169 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1170 const struct flow_keys *keys, int hlen);
1171 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1172 void *data, int hlen_proto);
1174 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1175 int thoff, u8 ip_proto)
1177 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1180 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1181 const struct flow_dissector_key *key,
1182 unsigned int key_count);
1184 bool __skb_flow_dissect(const struct sk_buff *skb,
1185 struct flow_dissector *flow_dissector,
1186 void *target_container,
1187 void *data, __be16 proto, int nhoff, int hlen,
1188 unsigned int flags);
1190 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1191 struct flow_dissector *flow_dissector,
1192 void *target_container, unsigned int flags)
1194 return __skb_flow_dissect(skb, flow_dissector, target_container,
1195 NULL, 0, 0, 0, flags);
1198 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1199 struct flow_keys *flow,
1202 memset(flow, 0, sizeof(*flow));
1203 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1204 NULL, 0, 0, 0, flags);
1207 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1208 void *data, __be16 proto,
1209 int nhoff, int hlen,
1212 memset(flow, 0, sizeof(*flow));
1213 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1214 data, proto, nhoff, hlen, flags);
1218 skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1219 struct flow_dissector *flow_dissector,
1220 void *target_container);
1222 static inline __u32 skb_get_hash(struct sk_buff *skb)
1224 if (!skb->l4_hash && !skb->sw_hash)
1225 __skb_get_hash(skb);
1230 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1232 if (!skb->l4_hash && !skb->sw_hash) {
1233 struct flow_keys keys;
1234 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1236 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1242 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1244 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1249 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1251 to->hash = from->hash;
1252 to->sw_hash = from->sw_hash;
1253 to->l4_hash = from->l4_hash;
1256 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1257 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1259 return skb->head + skb->end;
1262 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1267 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1272 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1274 return skb->end - skb->head;
1279 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1281 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1283 return &skb_shinfo(skb)->hwtstamps;
1286 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1288 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1290 return is_zcopy ? skb_uarg(skb) : NULL;
1293 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1295 if (skb && uarg && !skb_zcopy(skb)) {
1296 sock_zerocopy_get(uarg);
1297 skb_shinfo(skb)->destructor_arg = uarg;
1298 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1302 /* Release a reference on a zerocopy structure */
1303 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1305 struct ubuf_info *uarg = skb_zcopy(skb);
1308 if (uarg->callback == sock_zerocopy_callback) {
1309 uarg->zerocopy = uarg->zerocopy && zerocopy;
1310 sock_zerocopy_put(uarg);
1312 uarg->callback(uarg, zerocopy);
1315 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1319 /* Abort a zerocopy operation and revert zckey on error in send syscall */
1320 static inline void skb_zcopy_abort(struct sk_buff *skb)
1322 struct ubuf_info *uarg = skb_zcopy(skb);
1325 sock_zerocopy_put_abort(uarg);
1326 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1331 * skb_queue_empty - check if a queue is empty
1334 * Returns true if the queue is empty, false otherwise.
1336 static inline int skb_queue_empty(const struct sk_buff_head *list)
1338 return list->next == (const struct sk_buff *) list;
1342 * skb_queue_is_last - check if skb is the last entry in the queue
1346 * Returns true if @skb is the last buffer on the list.
1348 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1349 const struct sk_buff *skb)
1351 return skb->next == (const struct sk_buff *) list;
1355 * skb_queue_is_first - check if skb is the first entry in the queue
1359 * Returns true if @skb is the first buffer on the list.
1361 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1362 const struct sk_buff *skb)
1364 return skb->prev == (const struct sk_buff *) list;
1368 * skb_queue_next - return the next packet in the queue
1370 * @skb: current buffer
1372 * Return the next packet in @list after @skb. It is only valid to
1373 * call this if skb_queue_is_last() evaluates to false.
1375 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1376 const struct sk_buff *skb)
1378 /* This BUG_ON may seem severe, but if we just return then we
1379 * are going to dereference garbage.
1381 BUG_ON(skb_queue_is_last(list, skb));
1386 * skb_queue_prev - return the prev packet in the queue
1388 * @skb: current buffer
1390 * Return the prev packet in @list before @skb. It is only valid to
1391 * call this if skb_queue_is_first() evaluates to false.
1393 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1394 const struct sk_buff *skb)
1396 /* This BUG_ON may seem severe, but if we just return then we
1397 * are going to dereference garbage.
1399 BUG_ON(skb_queue_is_first(list, skb));
1404 * skb_get - reference buffer
1405 * @skb: buffer to reference
1407 * Makes another reference to a socket buffer and returns a pointer
1410 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1412 refcount_inc(&skb->users);
1417 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1421 * skb_cloned - is the buffer a clone
1422 * @skb: buffer to check
1424 * Returns true if the buffer was generated with skb_clone() and is
1425 * one of multiple shared copies of the buffer. Cloned buffers are
1426 * shared data so must not be written to under normal circumstances.
1428 static inline int skb_cloned(const struct sk_buff *skb)
1430 return skb->cloned &&
1431 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1434 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1436 might_sleep_if(gfpflags_allow_blocking(pri));
1438 if (skb_cloned(skb))
1439 return pskb_expand_head(skb, 0, 0, pri);
1445 * skb_header_cloned - is the header a clone
1446 * @skb: buffer to check
1448 * Returns true if modifying the header part of the buffer requires
1449 * the data to be copied.
1451 static inline int skb_header_cloned(const struct sk_buff *skb)
1458 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1459 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1460 return dataref != 1;
1463 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1465 might_sleep_if(gfpflags_allow_blocking(pri));
1467 if (skb_header_cloned(skb))
1468 return pskb_expand_head(skb, 0, 0, pri);
1474 * __skb_header_release - release reference to header
1475 * @skb: buffer to operate on
1477 static inline void __skb_header_release(struct sk_buff *skb)
1480 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1485 * skb_shared - is the buffer shared
1486 * @skb: buffer to check
1488 * Returns true if more than one person has a reference to this
1491 static inline int skb_shared(const struct sk_buff *skb)
1493 return refcount_read(&skb->users) != 1;
1497 * skb_share_check - check if buffer is shared and if so clone it
1498 * @skb: buffer to check
1499 * @pri: priority for memory allocation
1501 * If the buffer is shared the buffer is cloned and the old copy
1502 * drops a reference. A new clone with a single reference is returned.
1503 * If the buffer is not shared the original buffer is returned. When
1504 * being called from interrupt status or with spinlocks held pri must
1507 * NULL is returned on a memory allocation failure.
1509 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1511 might_sleep_if(gfpflags_allow_blocking(pri));
1512 if (skb_shared(skb)) {
1513 struct sk_buff *nskb = skb_clone(skb, pri);
1525 * Copy shared buffers into a new sk_buff. We effectively do COW on
1526 * packets to handle cases where we have a local reader and forward
1527 * and a couple of other messy ones. The normal one is tcpdumping
1528 * a packet thats being forwarded.
1532 * skb_unshare - make a copy of a shared buffer
1533 * @skb: buffer to check
1534 * @pri: priority for memory allocation
1536 * If the socket buffer is a clone then this function creates a new
1537 * copy of the data, drops a reference count on the old copy and returns
1538 * the new copy with the reference count at 1. If the buffer is not a clone
1539 * the original buffer is returned. When called with a spinlock held or
1540 * from interrupt state @pri must be %GFP_ATOMIC
1542 * %NULL is returned on a memory allocation failure.
1544 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1547 might_sleep_if(gfpflags_allow_blocking(pri));
1548 if (skb_cloned(skb)) {
1549 struct sk_buff *nskb = skb_copy(skb, pri);
1551 /* Free our shared copy */
1562 * skb_peek - peek at the head of an &sk_buff_head
1563 * @list_: list to peek at
1565 * Peek an &sk_buff. Unlike most other operations you _MUST_
1566 * be careful with this one. A peek leaves the buffer on the
1567 * list and someone else may run off with it. You must hold
1568 * the appropriate locks or have a private queue to do this.
1570 * Returns %NULL for an empty list or a pointer to the head element.
1571 * The reference count is not incremented and the reference is therefore
1572 * volatile. Use with caution.
1574 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1576 struct sk_buff *skb = list_->next;
1578 if (skb == (struct sk_buff *)list_)
1584 * skb_peek_next - peek skb following the given one from a queue
1585 * @skb: skb to start from
1586 * @list_: list to peek at
1588 * Returns %NULL when the end of the list is met or a pointer to the
1589 * next element. The reference count is not incremented and the
1590 * reference is therefore volatile. Use with caution.
1592 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1593 const struct sk_buff_head *list_)
1595 struct sk_buff *next = skb->next;
1597 if (next == (struct sk_buff *)list_)
1603 * skb_peek_tail - peek at the tail of an &sk_buff_head
1604 * @list_: list to peek at
1606 * Peek an &sk_buff. Unlike most other operations you _MUST_
1607 * be careful with this one. A peek leaves the buffer on the
1608 * list and someone else may run off with it. You must hold
1609 * the appropriate locks or have a private queue to do this.
1611 * Returns %NULL for an empty list or a pointer to the tail element.
1612 * The reference count is not incremented and the reference is therefore
1613 * volatile. Use with caution.
1615 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1617 struct sk_buff *skb = list_->prev;
1619 if (skb == (struct sk_buff *)list_)
1626 * skb_queue_len - get queue length
1627 * @list_: list to measure
1629 * Return the length of an &sk_buff queue.
1631 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1637 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1638 * @list: queue to initialize
1640 * This initializes only the list and queue length aspects of
1641 * an sk_buff_head object. This allows to initialize the list
1642 * aspects of an sk_buff_head without reinitializing things like
1643 * the spinlock. It can also be used for on-stack sk_buff_head
1644 * objects where the spinlock is known to not be used.
1646 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1648 list->prev = list->next = (struct sk_buff *)list;
1653 * This function creates a split out lock class for each invocation;
1654 * this is needed for now since a whole lot of users of the skb-queue
1655 * infrastructure in drivers have different locking usage (in hardirq)
1656 * than the networking core (in softirq only). In the long run either the
1657 * network layer or drivers should need annotation to consolidate the
1658 * main types of usage into 3 classes.
1660 static inline void skb_queue_head_init(struct sk_buff_head *list)
1662 spin_lock_init(&list->lock);
1663 __skb_queue_head_init(list);
1666 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1667 struct lock_class_key *class)
1669 skb_queue_head_init(list);
1670 lockdep_set_class(&list->lock, class);
1674 * Insert an sk_buff on a list.
1676 * The "__skb_xxxx()" functions are the non-atomic ones that
1677 * can only be called with interrupts disabled.
1679 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1680 struct sk_buff_head *list);
1681 static inline void __skb_insert(struct sk_buff *newsk,
1682 struct sk_buff *prev, struct sk_buff *next,
1683 struct sk_buff_head *list)
1687 next->prev = prev->next = newsk;
1691 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1692 struct sk_buff *prev,
1693 struct sk_buff *next)
1695 struct sk_buff *first = list->next;
1696 struct sk_buff *last = list->prev;
1706 * skb_queue_splice - join two skb lists, this is designed for stacks
1707 * @list: the new list to add
1708 * @head: the place to add it in the first list
1710 static inline void skb_queue_splice(const struct sk_buff_head *list,
1711 struct sk_buff_head *head)
1713 if (!skb_queue_empty(list)) {
1714 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1715 head->qlen += list->qlen;
1720 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1721 * @list: the new list to add
1722 * @head: the place to add it in the first list
1724 * The list at @list is reinitialised
1726 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1727 struct sk_buff_head *head)
1729 if (!skb_queue_empty(list)) {
1730 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1731 head->qlen += list->qlen;
1732 __skb_queue_head_init(list);
1737 * skb_queue_splice_tail - join two skb lists, each list being a queue
1738 * @list: the new list to add
1739 * @head: the place to add it in the first list
1741 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1742 struct sk_buff_head *head)
1744 if (!skb_queue_empty(list)) {
1745 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1746 head->qlen += list->qlen;
1751 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1752 * @list: the new list to add
1753 * @head: the place to add it in the first list
1755 * Each of the lists is a queue.
1756 * The list at @list is reinitialised
1758 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1759 struct sk_buff_head *head)
1761 if (!skb_queue_empty(list)) {
1762 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1763 head->qlen += list->qlen;
1764 __skb_queue_head_init(list);
1769 * __skb_queue_after - queue a buffer at the list head
1770 * @list: list to use
1771 * @prev: place after this buffer
1772 * @newsk: buffer to queue
1774 * Queue a buffer int the middle of a list. This function takes no locks
1775 * and you must therefore hold required locks before calling it.
1777 * A buffer cannot be placed on two lists at the same time.
1779 static inline void __skb_queue_after(struct sk_buff_head *list,
1780 struct sk_buff *prev,
1781 struct sk_buff *newsk)
1783 __skb_insert(newsk, prev, prev->next, list);
1786 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1787 struct sk_buff_head *list);
1789 static inline void __skb_queue_before(struct sk_buff_head *list,
1790 struct sk_buff *next,
1791 struct sk_buff *newsk)
1793 __skb_insert(newsk, next->prev, next, list);
1797 * __skb_queue_head - queue a buffer at the list head
1798 * @list: list to use
1799 * @newsk: buffer to queue
1801 * Queue a buffer at the start of a list. This function takes no locks
1802 * and you must therefore hold required locks before calling it.
1804 * A buffer cannot be placed on two lists at the same time.
1806 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1807 static inline void __skb_queue_head(struct sk_buff_head *list,
1808 struct sk_buff *newsk)
1810 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1814 * __skb_queue_tail - queue a buffer at the list tail
1815 * @list: list to use
1816 * @newsk: buffer to queue
1818 * Queue a buffer at the end of a list. This function takes no locks
1819 * and you must therefore hold required locks before calling it.
1821 * A buffer cannot be placed on two lists at the same time.
1823 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1824 static inline void __skb_queue_tail(struct sk_buff_head *list,
1825 struct sk_buff *newsk)
1827 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1831 * remove sk_buff from list. _Must_ be called atomically, and with
1834 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1835 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1837 struct sk_buff *next, *prev;
1842 skb->next = skb->prev = NULL;
1848 * __skb_dequeue - remove from the head of the queue
1849 * @list: list to dequeue from
1851 * Remove the head of the list. This function does not take any locks
1852 * so must be used with appropriate locks held only. The head item is
1853 * returned or %NULL if the list is empty.
1855 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1856 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1858 struct sk_buff *skb = skb_peek(list);
1860 __skb_unlink(skb, list);
1865 * __skb_dequeue_tail - remove from the tail of the queue
1866 * @list: list to dequeue from
1868 * Remove the tail of the list. This function does not take any locks
1869 * so must be used with appropriate locks held only. The tail item is
1870 * returned or %NULL if the list is empty.
1872 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1873 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1875 struct sk_buff *skb = skb_peek_tail(list);
1877 __skb_unlink(skb, list);
1882 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1884 return skb->data_len;
1887 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1889 return skb->len - skb->data_len;
1892 static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
1894 unsigned int i, len = 0;
1896 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1897 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1901 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1903 return skb_headlen(skb) + __skb_pagelen(skb);
1907 * __skb_fill_page_desc - initialise a paged fragment in an skb
1908 * @skb: buffer containing fragment to be initialised
1909 * @i: paged fragment index to initialise
1910 * @page: the page to use for this fragment
1911 * @off: the offset to the data with @page
1912 * @size: the length of the data
1914 * Initialises the @i'th fragment of @skb to point to &size bytes at
1915 * offset @off within @page.
1917 * Does not take any additional reference on the fragment.
1919 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1920 struct page *page, int off, int size)
1922 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1925 * Propagate page pfmemalloc to the skb if we can. The problem is
1926 * that not all callers have unique ownership of the page but rely
1927 * on page_is_pfmemalloc doing the right thing(tm).
1929 frag->page.p = page;
1930 frag->page_offset = off;
1931 skb_frag_size_set(frag, size);
1933 page = compound_head(page);
1934 if (page_is_pfmemalloc(page))
1935 skb->pfmemalloc = true;
1939 * skb_fill_page_desc - initialise a paged fragment in an skb
1940 * @skb: buffer containing fragment to be initialised
1941 * @i: paged fragment index to initialise
1942 * @page: the page to use for this fragment
1943 * @off: the offset to the data with @page
1944 * @size: the length of the data
1946 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1947 * @skb to point to @size bytes at offset @off within @page. In
1948 * addition updates @skb such that @i is the last fragment.
1950 * Does not take any additional reference on the fragment.
1952 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1953 struct page *page, int off, int size)
1955 __skb_fill_page_desc(skb, i, page, off, size);
1956 skb_shinfo(skb)->nr_frags = i + 1;
1959 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1960 int size, unsigned int truesize);
1962 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1963 unsigned int truesize);
1965 #define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
1966 #define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1967 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1969 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1970 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1972 return skb->head + skb->tail;
1975 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1977 skb->tail = skb->data - skb->head;
1980 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1982 skb_reset_tail_pointer(skb);
1983 skb->tail += offset;
1986 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1987 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1992 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1994 skb->tail = skb->data;
1997 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1999 skb->tail = skb->data + offset;
2002 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
2005 * Add data to an sk_buff
2007 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2008 void *skb_put(struct sk_buff *skb, unsigned int len);
2009 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2011 void *tmp = skb_tail_pointer(skb);
2012 SKB_LINEAR_ASSERT(skb);
2018 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2020 void *tmp = __skb_put(skb, len);
2022 memset(tmp, 0, len);
2026 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2029 void *tmp = __skb_put(skb, len);
2031 memcpy(tmp, data, len);
2035 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2037 *(u8 *)__skb_put(skb, 1) = val;
2040 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2042 void *tmp = skb_put(skb, len);
2044 memset(tmp, 0, len);
2049 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2052 void *tmp = skb_put(skb, len);
2054 memcpy(tmp, data, len);
2059 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2061 *(u8 *)skb_put(skb, 1) = val;
2064 void *skb_push(struct sk_buff *skb, unsigned int len);
2065 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2072 void *skb_pull(struct sk_buff *skb, unsigned int len);
2073 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2076 BUG_ON(skb->len < skb->data_len);
2077 return skb->data += len;
2080 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2082 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2085 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2087 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2089 if (len > skb_headlen(skb) &&
2090 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2093 return skb->data += len;
2096 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2098 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2101 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2103 if (likely(len <= skb_headlen(skb)))
2105 if (unlikely(len > skb->len))
2107 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2110 void skb_condense(struct sk_buff *skb);
2113 * skb_headroom - bytes at buffer head
2114 * @skb: buffer to check
2116 * Return the number of bytes of free space at the head of an &sk_buff.
2118 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2120 return skb->data - skb->head;
2124 * skb_tailroom - bytes at buffer end
2125 * @skb: buffer to check
2127 * Return the number of bytes of free space at the tail of an sk_buff
2129 static inline int skb_tailroom(const struct sk_buff *skb)
2131 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2135 * skb_availroom - bytes at buffer end
2136 * @skb: buffer to check
2138 * Return the number of bytes of free space at the tail of an sk_buff
2139 * allocated by sk_stream_alloc()
2141 static inline int skb_availroom(const struct sk_buff *skb)
2143 if (skb_is_nonlinear(skb))
2146 return skb->end - skb->tail - skb->reserved_tailroom;
2150 * skb_reserve - adjust headroom
2151 * @skb: buffer to alter
2152 * @len: bytes to move
2154 * Increase the headroom of an empty &sk_buff by reducing the tail
2155 * room. This is only allowed for an empty buffer.
2157 static inline void skb_reserve(struct sk_buff *skb, int len)
2164 * skb_tailroom_reserve - adjust reserved_tailroom
2165 * @skb: buffer to alter
2166 * @mtu: maximum amount of headlen permitted
2167 * @needed_tailroom: minimum amount of reserved_tailroom
2169 * Set reserved_tailroom so that headlen can be as large as possible but
2170 * not larger than mtu and tailroom cannot be smaller than
2172 * The required headroom should already have been reserved before using
2175 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2176 unsigned int needed_tailroom)
2178 SKB_LINEAR_ASSERT(skb);
2179 if (mtu < skb_tailroom(skb) - needed_tailroom)
2180 /* use at most mtu */
2181 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2183 /* use up to all available space */
2184 skb->reserved_tailroom = needed_tailroom;
2187 #define ENCAP_TYPE_ETHER 0
2188 #define ENCAP_TYPE_IPPROTO 1
2190 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2193 skb->inner_protocol = protocol;
2194 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2197 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2200 skb->inner_ipproto = ipproto;
2201 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2204 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2206 skb->inner_mac_header = skb->mac_header;
2207 skb->inner_network_header = skb->network_header;
2208 skb->inner_transport_header = skb->transport_header;
2211 static inline void skb_reset_mac_len(struct sk_buff *skb)
2213 skb->mac_len = skb->network_header - skb->mac_header;
2216 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2219 return skb->head + skb->inner_transport_header;
2222 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2224 return skb_inner_transport_header(skb) - skb->data;
2227 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2229 skb->inner_transport_header = skb->data - skb->head;
2232 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2235 skb_reset_inner_transport_header(skb);
2236 skb->inner_transport_header += offset;
2239 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2241 return skb->head + skb->inner_network_header;
2244 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2246 skb->inner_network_header = skb->data - skb->head;
2249 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2252 skb_reset_inner_network_header(skb);
2253 skb->inner_network_header += offset;
2256 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2258 return skb->head + skb->inner_mac_header;
2261 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2263 skb->inner_mac_header = skb->data - skb->head;
2266 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2269 skb_reset_inner_mac_header(skb);
2270 skb->inner_mac_header += offset;
2272 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2274 return skb->transport_header != (typeof(skb->transport_header))~0U;
2277 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2279 return skb->head + skb->transport_header;
2282 static inline void skb_reset_transport_header(struct sk_buff *skb)
2284 skb->transport_header = skb->data - skb->head;
2287 static inline void skb_set_transport_header(struct sk_buff *skb,
2290 skb_reset_transport_header(skb);
2291 skb->transport_header += offset;
2294 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2296 return skb->head + skb->network_header;
2299 static inline void skb_reset_network_header(struct sk_buff *skb)
2301 skb->network_header = skb->data - skb->head;
2304 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2306 skb_reset_network_header(skb);
2307 skb->network_header += offset;
2310 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2312 return skb->head + skb->mac_header;
2315 static inline int skb_mac_offset(const struct sk_buff *skb)
2317 return skb_mac_header(skb) - skb->data;
2320 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2322 return skb->network_header - skb->mac_header;
2325 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2327 return skb->mac_header != (typeof(skb->mac_header))~0U;
2330 static inline void skb_reset_mac_header(struct sk_buff *skb)
2332 skb->mac_header = skb->data - skb->head;
2335 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2337 skb_reset_mac_header(skb);
2338 skb->mac_header += offset;
2341 static inline void skb_pop_mac_header(struct sk_buff *skb)
2343 skb->mac_header = skb->network_header;
2346 static inline void skb_probe_transport_header(struct sk_buff *skb,
2347 const int offset_hint)
2349 struct flow_keys keys;
2351 if (skb_transport_header_was_set(skb))
2353 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2354 skb_set_transport_header(skb, keys.control.thoff);
2356 skb_set_transport_header(skb, offset_hint);
2359 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2361 if (skb_mac_header_was_set(skb)) {
2362 const unsigned char *old_mac = skb_mac_header(skb);
2364 skb_set_mac_header(skb, -skb->mac_len);
2365 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2369 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2371 return skb->csum_start - skb_headroom(skb);
2374 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2376 return skb->head + skb->csum_start;
2379 static inline int skb_transport_offset(const struct sk_buff *skb)
2381 return skb_transport_header(skb) - skb->data;
2384 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2386 return skb->transport_header - skb->network_header;
2389 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2391 return skb->inner_transport_header - skb->inner_network_header;
2394 static inline int skb_network_offset(const struct sk_buff *skb)
2396 return skb_network_header(skb) - skb->data;
2399 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2401 return skb_inner_network_header(skb) - skb->data;
2404 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2406 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2410 * CPUs often take a performance hit when accessing unaligned memory
2411 * locations. The actual performance hit varies, it can be small if the
2412 * hardware handles it or large if we have to take an exception and fix it
2415 * Since an ethernet header is 14 bytes network drivers often end up with
2416 * the IP header at an unaligned offset. The IP header can be aligned by
2417 * shifting the start of the packet by 2 bytes. Drivers should do this
2420 * skb_reserve(skb, NET_IP_ALIGN);
2422 * The downside to this alignment of the IP header is that the DMA is now
2423 * unaligned. On some architectures the cost of an unaligned DMA is high
2424 * and this cost outweighs the gains made by aligning the IP header.
2426 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2429 #ifndef NET_IP_ALIGN
2430 #define NET_IP_ALIGN 2
2434 * The networking layer reserves some headroom in skb data (via
2435 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2436 * the header has to grow. In the default case, if the header has to grow
2437 * 32 bytes or less we avoid the reallocation.
2439 * Unfortunately this headroom changes the DMA alignment of the resulting
2440 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2441 * on some architectures. An architecture can override this value,
2442 * perhaps setting it to a cacheline in size (since that will maintain
2443 * cacheline alignment of the DMA). It must be a power of 2.
2445 * Various parts of the networking layer expect at least 32 bytes of
2446 * headroom, you should not reduce this.
2448 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2449 * to reduce average number of cache lines per packet.
2450 * get_rps_cpus() for example only access one 64 bytes aligned block :
2451 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2454 #define NET_SKB_PAD max(32, L1_CACHE_BYTES)
2457 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2459 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2461 if (unlikely(skb_is_nonlinear(skb))) {
2466 skb_set_tail_pointer(skb, len);
2469 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2471 __skb_set_length(skb, len);
2474 void skb_trim(struct sk_buff *skb, unsigned int len);
2476 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2479 return ___pskb_trim(skb, len);
2480 __skb_trim(skb, len);
2484 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2486 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2490 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2491 * @skb: buffer to alter
2494 * This is identical to pskb_trim except that the caller knows that
2495 * the skb is not cloned so we should never get an error due to out-
2498 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2500 int err = pskb_trim(skb, len);
2504 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2506 unsigned int diff = len - skb->len;
2508 if (skb_tailroom(skb) < diff) {
2509 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2514 __skb_set_length(skb, len);
2519 * skb_orphan - orphan a buffer
2520 * @skb: buffer to orphan
2522 * If a buffer currently has an owner then we call the owner's
2523 * destructor function and make the @skb unowned. The buffer continues
2524 * to exist but is no longer charged to its former owner.
2526 static inline void skb_orphan(struct sk_buff *skb)
2528 if (skb->destructor) {
2529 skb->destructor(skb);
2530 skb->destructor = NULL;
2538 * skb_orphan_frags - orphan the frags contained in a buffer
2539 * @skb: buffer to orphan frags from
2540 * @gfp_mask: allocation mask for replacement pages
2542 * For each frag in the SKB which needs a destructor (i.e. has an
2543 * owner) create a copy of that frag and release the original
2544 * page by calling the destructor.
2546 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2548 if (likely(!skb_zcopy(skb)))
2550 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2552 return skb_copy_ubufs(skb, gfp_mask);
2555 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2556 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2558 if (likely(!skb_zcopy(skb)))
2560 return skb_copy_ubufs(skb, gfp_mask);
2564 * __skb_queue_purge - empty a list
2565 * @list: list to empty
2567 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2568 * the list and one reference dropped. This function does not take the
2569 * list lock and the caller must hold the relevant locks to use it.
2571 void skb_queue_purge(struct sk_buff_head *list);
2572 static inline void __skb_queue_purge(struct sk_buff_head *list)
2574 struct sk_buff *skb;
2575 while ((skb = __skb_dequeue(list)) != NULL)
2579 void skb_rbtree_purge(struct rb_root *root);
2581 void *netdev_alloc_frag(unsigned int fragsz);
2583 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2587 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2588 * @dev: network device to receive on
2589 * @length: length to allocate
2591 * Allocate a new &sk_buff and assign it a usage count of one. The
2592 * buffer has unspecified headroom built in. Users should allocate
2593 * the headroom they think they need without accounting for the
2594 * built in space. The built in space is used for optimisations.
2596 * %NULL is returned if there is no free memory. Although this function
2597 * allocates memory it can be called from an interrupt.
2599 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2600 unsigned int length)
2602 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2605 /* legacy helper around __netdev_alloc_skb() */
2606 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2609 return __netdev_alloc_skb(NULL, length, gfp_mask);
2612 /* legacy helper around netdev_alloc_skb() */
2613 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2615 return netdev_alloc_skb(NULL, length);
2619 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2620 unsigned int length, gfp_t gfp)
2622 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2624 if (NET_IP_ALIGN && skb)
2625 skb_reserve(skb, NET_IP_ALIGN);
2629 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2630 unsigned int length)
2632 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2635 static inline void skb_free_frag(void *addr)
2637 page_frag_free(addr);
2640 void *napi_alloc_frag(unsigned int fragsz);
2641 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2642 unsigned int length, gfp_t gfp_mask);
2643 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2644 unsigned int length)
2646 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2648 void napi_consume_skb(struct sk_buff *skb, int budget);
2650 void __kfree_skb_flush(void);
2651 void __kfree_skb_defer(struct sk_buff *skb);
2654 * __dev_alloc_pages - allocate page for network Rx
2655 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2656 * @order: size of the allocation
2658 * Allocate a new page.
2660 * %NULL is returned if there is no free memory.
2662 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2665 /* This piece of code contains several assumptions.
2666 * 1. This is for device Rx, therefor a cold page is preferred.
2667 * 2. The expectation is the user wants a compound page.
2668 * 3. If requesting a order 0 page it will not be compound
2669 * due to the check to see if order has a value in prep_new_page
2670 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2671 * code in gfp_to_alloc_flags that should be enforcing this.
2673 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2675 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2678 static inline struct page *dev_alloc_pages(unsigned int order)
2680 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2684 * __dev_alloc_page - allocate a page for network Rx
2685 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2687 * Allocate a new page.
2689 * %NULL is returned if there is no free memory.
2691 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2693 return __dev_alloc_pages(gfp_mask, 0);
2696 static inline struct page *dev_alloc_page(void)
2698 return dev_alloc_pages(0);
2702 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2703 * @page: The page that was allocated from skb_alloc_page
2704 * @skb: The skb that may need pfmemalloc set
2706 static inline void skb_propagate_pfmemalloc(struct page *page,
2707 struct sk_buff *skb)
2709 if (page_is_pfmemalloc(page))
2710 skb->pfmemalloc = true;
2714 * skb_frag_page - retrieve the page referred to by a paged fragment
2715 * @frag: the paged fragment
2717 * Returns the &struct page associated with @frag.
2719 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2721 return frag->page.p;
2725 * __skb_frag_ref - take an addition reference on a paged fragment.
2726 * @frag: the paged fragment
2728 * Takes an additional reference on the paged fragment @frag.
2730 static inline void __skb_frag_ref(skb_frag_t *frag)
2732 get_page(skb_frag_page(frag));
2736 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2738 * @f: the fragment offset.
2740 * Takes an additional reference on the @f'th paged fragment of @skb.
2742 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2744 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2748 * __skb_frag_unref - release a reference on a paged fragment.
2749 * @frag: the paged fragment
2751 * Releases a reference on the paged fragment @frag.
2753 static inline void __skb_frag_unref(skb_frag_t *frag)
2755 put_page(skb_frag_page(frag));
2759 * skb_frag_unref - release a reference on a paged fragment of an skb.
2761 * @f: the fragment offset
2763 * Releases a reference on the @f'th paged fragment of @skb.
2765 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2767 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2771 * skb_frag_address - gets the address of the data contained in a paged fragment
2772 * @frag: the paged fragment buffer
2774 * Returns the address of the data within @frag. The page must already
2777 static inline void *skb_frag_address(const skb_frag_t *frag)
2779 return page_address(skb_frag_page(frag)) + frag->page_offset;
2783 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2784 * @frag: the paged fragment buffer
2786 * Returns the address of the data within @frag. Checks that the page
2787 * is mapped and returns %NULL otherwise.
2789 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2791 void *ptr = page_address(skb_frag_page(frag));
2795 return ptr + frag->page_offset;
2799 * __skb_frag_set_page - sets the page contained in a paged fragment
2800 * @frag: the paged fragment
2801 * @page: the page to set
2803 * Sets the fragment @frag to contain @page.
2805 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2807 frag->page.p = page;
2811 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2813 * @f: the fragment offset
2814 * @page: the page to set
2816 * Sets the @f'th fragment of @skb to contain @page.
2818 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2821 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2824 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2827 * skb_frag_dma_map - maps a paged fragment via the DMA API
2828 * @dev: the device to map the fragment to
2829 * @frag: the paged fragment to map
2830 * @offset: the offset within the fragment (starting at the
2831 * fragment's own offset)
2832 * @size: the number of bytes to map
2833 * @dir: the direction of the mapping (``PCI_DMA_*``)
2835 * Maps the page associated with @frag to @device.
2837 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2838 const skb_frag_t *frag,
2839 size_t offset, size_t size,
2840 enum dma_data_direction dir)
2842 return dma_map_page(dev, skb_frag_page(frag),
2843 frag->page_offset + offset, size, dir);
2846 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2849 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2853 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2856 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2861 * skb_clone_writable - is the header of a clone writable
2862 * @skb: buffer to check
2863 * @len: length up to which to write
2865 * Returns true if modifying the header part of the cloned buffer
2866 * does not requires the data to be copied.
2868 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2870 return !skb_header_cloned(skb) &&
2871 skb_headroom(skb) + len <= skb->hdr_len;
2874 static inline int skb_try_make_writable(struct sk_buff *skb,
2875 unsigned int write_len)
2877 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2878 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2881 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2886 if (headroom > skb_headroom(skb))
2887 delta = headroom - skb_headroom(skb);
2889 if (delta || cloned)
2890 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2896 * skb_cow - copy header of skb when it is required
2897 * @skb: buffer to cow
2898 * @headroom: needed headroom
2900 * If the skb passed lacks sufficient headroom or its data part
2901 * is shared, data is reallocated. If reallocation fails, an error
2902 * is returned and original skb is not changed.
2904 * The result is skb with writable area skb->head...skb->tail
2905 * and at least @headroom of space at head.
2907 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2909 return __skb_cow(skb, headroom, skb_cloned(skb));
2913 * skb_cow_head - skb_cow but only making the head writable
2914 * @skb: buffer to cow
2915 * @headroom: needed headroom
2917 * This function is identical to skb_cow except that we replace the
2918 * skb_cloned check by skb_header_cloned. It should be used when
2919 * you only need to push on some header and do not need to modify
2922 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2924 return __skb_cow(skb, headroom, skb_header_cloned(skb));
2928 * skb_padto - pad an skbuff up to a minimal size
2929 * @skb: buffer to pad
2930 * @len: minimal length
2932 * Pads up a buffer to ensure the trailing bytes exist and are
2933 * blanked. If the buffer already contains sufficient data it
2934 * is untouched. Otherwise it is extended. Returns zero on
2935 * success. The skb is freed on error.
2937 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2939 unsigned int size = skb->len;
2940 if (likely(size >= len))
2942 return skb_pad(skb, len - size);
2946 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2947 * @skb: buffer to pad
2948 * @len: minimal length
2949 * @free_on_error: free buffer on error
2951 * Pads up a buffer to ensure the trailing bytes exist and are
2952 * blanked. If the buffer already contains sufficient data it
2953 * is untouched. Otherwise it is extended. Returns zero on
2954 * success. The skb is freed on error if @free_on_error is true.
2956 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2959 unsigned int size = skb->len;
2961 if (unlikely(size < len)) {
2963 if (__skb_pad(skb, len, free_on_error))
2965 __skb_put(skb, len);
2971 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2972 * @skb: buffer to pad
2973 * @len: minimal length
2975 * Pads up a buffer to ensure the trailing bytes exist and are
2976 * blanked. If the buffer already contains sufficient data it
2977 * is untouched. Otherwise it is extended. Returns zero on
2978 * success. The skb is freed on error.
2980 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2982 return __skb_put_padto(skb, len, true);
2985 static inline int skb_add_data(struct sk_buff *skb,
2986 struct iov_iter *from, int copy)
2988 const int off = skb->len;
2990 if (skb->ip_summed == CHECKSUM_NONE) {
2992 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2994 skb->csum = csum_block_add(skb->csum, csum, off);
2997 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3000 __skb_trim(skb, off);
3004 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3005 const struct page *page, int off)
3010 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3012 return page == skb_frag_page(frag) &&
3013 off == frag->page_offset + skb_frag_size(frag);
3018 static inline int __skb_linearize(struct sk_buff *skb)
3020 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3024 * skb_linearize - convert paged skb to linear one
3025 * @skb: buffer to linarize
3027 * If there is no free memory -ENOMEM is returned, otherwise zero
3028 * is returned and the old skb data released.
3030 static inline int skb_linearize(struct sk_buff *skb)
3032 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3036 * skb_has_shared_frag - can any frag be overwritten
3037 * @skb: buffer to test
3039 * Return true if the skb has at least one frag that might be modified
3040 * by an external entity (as in vmsplice()/sendfile())
3042 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3044 return skb_is_nonlinear(skb) &&
3045 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3049 * skb_linearize_cow - make sure skb is linear and writable
3050 * @skb: buffer to process
3052 * If there is no free memory -ENOMEM is returned, otherwise zero
3053 * is returned and the old skb data released.
3055 static inline int skb_linearize_cow(struct sk_buff *skb)
3057 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3058 __skb_linearize(skb) : 0;
3061 static __always_inline void
3062 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3065 if (skb->ip_summed == CHECKSUM_COMPLETE)
3066 skb->csum = csum_block_sub(skb->csum,
3067 csum_partial(start, len, 0), off);
3068 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3069 skb_checksum_start_offset(skb) < 0)
3070 skb->ip_summed = CHECKSUM_NONE;
3074 * skb_postpull_rcsum - update checksum for received skb after pull
3075 * @skb: buffer to update
3076 * @start: start of data before pull
3077 * @len: length of data pulled
3079 * After doing a pull on a received packet, you need to call this to
3080 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3081 * CHECKSUM_NONE so that it can be recomputed from scratch.
3083 static inline void skb_postpull_rcsum(struct sk_buff *skb,
3084 const void *start, unsigned int len)
3086 __skb_postpull_rcsum(skb, start, len, 0);
3089 static __always_inline void
3090 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3093 if (skb->ip_summed == CHECKSUM_COMPLETE)
3094 skb->csum = csum_block_add(skb->csum,
3095 csum_partial(start, len, 0), off);
3099 * skb_postpush_rcsum - update checksum for received skb after push
3100 * @skb: buffer to update
3101 * @start: start of data after push
3102 * @len: length of data pushed
3104 * After doing a push on a received packet, you need to call this to
3105 * update the CHECKSUM_COMPLETE checksum.
3107 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3108 const void *start, unsigned int len)
3110 __skb_postpush_rcsum(skb, start, len, 0);
3113 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3116 * skb_push_rcsum - push skb and update receive checksum
3117 * @skb: buffer to update
3118 * @len: length of data pulled
3120 * This function performs an skb_push on the packet and updates
3121 * the CHECKSUM_COMPLETE checksum. It should be used on
3122 * receive path processing instead of skb_push unless you know
3123 * that the checksum difference is zero (e.g., a valid IP header)
3124 * or you are setting ip_summed to CHECKSUM_NONE.
3126 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3129 skb_postpush_rcsum(skb, skb->data, len);
3134 * pskb_trim_rcsum - trim received skb and update checksum
3135 * @skb: buffer to trim
3138 * This is exactly the same as pskb_trim except that it ensures the
3139 * checksum of received packets are still valid after the operation.
3142 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3144 if (likely(len >= skb->len))
3146 if (skb->ip_summed == CHECKSUM_COMPLETE)
3147 skb->ip_summed = CHECKSUM_NONE;
3148 return __pskb_trim(skb, len);
3151 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3153 if (skb->ip_summed == CHECKSUM_COMPLETE)
3154 skb->ip_summed = CHECKSUM_NONE;
3155 __skb_trim(skb, len);
3159 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3161 if (skb->ip_summed == CHECKSUM_COMPLETE)
3162 skb->ip_summed = CHECKSUM_NONE;
3163 return __skb_grow(skb, len);
3166 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3167 #define skb_rb_first(root) rb_to_skb(rb_first(root))
3168 #define skb_rb_last(root) rb_to_skb(rb_last(root))
3169 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode))
3170 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode))
3172 #define skb_queue_walk(queue, skb) \
3173 for (skb = (queue)->next; \
3174 skb != (struct sk_buff *)(queue); \
3177 #define skb_queue_walk_safe(queue, skb, tmp) \
3178 for (skb = (queue)->next, tmp = skb->next; \
3179 skb != (struct sk_buff *)(queue); \
3180 skb = tmp, tmp = skb->next)
3182 #define skb_queue_walk_from(queue, skb) \
3183 for (; skb != (struct sk_buff *)(queue); \
3186 #define skb_rbtree_walk(skb, root) \
3187 for (skb = skb_rb_first(root); skb != NULL; \
3188 skb = skb_rb_next(skb))
3190 #define skb_rbtree_walk_from(skb) \
3191 for (; skb != NULL; \
3192 skb = skb_rb_next(skb))
3194 #define skb_rbtree_walk_from_safe(skb, tmp) \
3195 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \
3198 #define skb_queue_walk_from_safe(queue, skb, tmp) \
3199 for (tmp = skb->next; \
3200 skb != (struct sk_buff *)(queue); \
3201 skb = tmp, tmp = skb->next)
3203 #define skb_queue_reverse_walk(queue, skb) \
3204 for (skb = (queue)->prev; \
3205 skb != (struct sk_buff *)(queue); \
3208 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3209 for (skb = (queue)->prev, tmp = skb->prev; \
3210 skb != (struct sk_buff *)(queue); \
3211 skb = tmp, tmp = skb->prev)
3213 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3214 for (tmp = skb->prev; \
3215 skb != (struct sk_buff *)(queue); \
3216 skb = tmp, tmp = skb->prev)
3218 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3220 return skb_shinfo(skb)->frag_list != NULL;
3223 static inline void skb_frag_list_init(struct sk_buff *skb)
3225 skb_shinfo(skb)->frag_list = NULL;
3228 #define skb_walk_frags(skb, iter) \
3229 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3232 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3233 const struct sk_buff *skb);
3234 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3235 struct sk_buff_head *queue,
3237 void (*destructor)(struct sock *sk,
3238 struct sk_buff *skb),
3239 int *peeked, int *off, int *err,
3240 struct sk_buff **last);
3241 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3242 void (*destructor)(struct sock *sk,
3243 struct sk_buff *skb),
3244 int *peeked, int *off, int *err,
3245 struct sk_buff **last);
3246 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3247 void (*destructor)(struct sock *sk,
3248 struct sk_buff *skb),
3249 int *peeked, int *off, int *err);
3250 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3252 __poll_t datagram_poll(struct file *file, struct socket *sock,
3253 struct poll_table_struct *wait);
3254 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3255 struct iov_iter *to, int size);
3256 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3257 struct msghdr *msg, int size)
3259 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3261 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3262 struct msghdr *msg);
3263 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3264 struct iov_iter *from, int len);
3265 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3266 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3267 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3268 static inline void skb_free_datagram_locked(struct sock *sk,
3269 struct sk_buff *skb)
3271 __skb_free_datagram_locked(sk, skb, 0);
3273 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3274 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3275 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3276 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3277 int len, __wsum csum);
3278 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3279 struct pipe_inode_info *pipe, unsigned int len,
3280 unsigned int flags);
3281 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3283 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3284 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3285 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3286 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3288 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3289 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3290 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3291 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3292 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3293 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3294 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3295 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3296 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3297 int skb_vlan_pop(struct sk_buff *skb);
3298 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3299 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3302 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3304 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3307 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3309 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3312 struct skb_checksum_ops {
3313 __wsum (*update)(const void *mem, int len, __wsum wsum);
3314 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3317 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3319 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3320 __wsum csum, const struct skb_checksum_ops *ops);
3321 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3324 static inline void * __must_check
3325 __skb_header_pointer(const struct sk_buff *skb, int offset,
3326 int len, void *data, int hlen, void *buffer)
3328 if (hlen - offset >= len)
3329 return data + offset;
3332 skb_copy_bits(skb, offset, buffer, len) < 0)
3338 static inline void * __must_check
3339 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3341 return __skb_header_pointer(skb, offset, len, skb->data,
3342 skb_headlen(skb), buffer);
3346 * skb_needs_linearize - check if we need to linearize a given skb
3347 * depending on the given device features.
3348 * @skb: socket buffer to check
3349 * @features: net device features
3351 * Returns true if either:
3352 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3353 * 2. skb is fragmented and the device does not support SG.
3355 static inline bool skb_needs_linearize(struct sk_buff *skb,
3356 netdev_features_t features)
3358 return skb_is_nonlinear(skb) &&
3359 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3360 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3363 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3365 const unsigned int len)
3367 memcpy(to, skb->data, len);
3370 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3371 const int offset, void *to,
3372 const unsigned int len)
3374 memcpy(to, skb->data + offset, len);
3377 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3379 const unsigned int len)
3381 memcpy(skb->data, from, len);
3384 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3387 const unsigned int len)
3389 memcpy(skb->data + offset, from, len);
3392 void skb_init(void);
3394 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3400 * skb_get_timestamp - get timestamp from a skb
3401 * @skb: skb to get stamp from
3402 * @stamp: pointer to struct timeval to store stamp in
3404 * Timestamps are stored in the skb as offsets to a base timestamp.
3405 * This function converts the offset back to a struct timeval and stores
3408 static inline void skb_get_timestamp(const struct sk_buff *skb,
3409 struct timeval *stamp)
3411 *stamp = ktime_to_timeval(skb->tstamp);
3414 static inline void skb_get_timestampns(const struct sk_buff *skb,
3415 struct timespec *stamp)
3417 *stamp = ktime_to_timespec(skb->tstamp);
3420 static inline void __net_timestamp(struct sk_buff *skb)
3422 skb->tstamp = ktime_get_real();
3425 static inline ktime_t net_timedelta(ktime_t t)
3427 return ktime_sub(ktime_get_real(), t);
3430 static inline ktime_t net_invalid_timestamp(void)
3435 static inline u8 skb_metadata_len(const struct sk_buff *skb)
3437 return skb_shinfo(skb)->meta_len;
3440 static inline void *skb_metadata_end(const struct sk_buff *skb)
3442 return skb_mac_header(skb);
3445 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3446 const struct sk_buff *skb_b,
3449 const void *a = skb_metadata_end(skb_a);
3450 const void *b = skb_metadata_end(skb_b);
3451 /* Using more efficient varaiant than plain call to memcmp(). */
3452 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3456 #define __it(x, op) (x -= sizeof(u##op))
3457 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3458 case 32: diffs |= __it_diff(a, b, 64);
3459 case 24: diffs |= __it_diff(a, b, 64);
3460 case 16: diffs |= __it_diff(a, b, 64);
3461 case 8: diffs |= __it_diff(a, b, 64);
3463 case 28: diffs |= __it_diff(a, b, 64);
3464 case 20: diffs |= __it_diff(a, b, 64);
3465 case 12: diffs |= __it_diff(a, b, 64);
3466 case 4: diffs |= __it_diff(a, b, 32);
3471 return memcmp(a - meta_len, b - meta_len, meta_len);
3475 static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3476 const struct sk_buff *skb_b)
3478 u8 len_a = skb_metadata_len(skb_a);
3479 u8 len_b = skb_metadata_len(skb_b);
3481 if (!(len_a | len_b))
3484 return len_a != len_b ?
3485 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3488 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3490 skb_shinfo(skb)->meta_len = meta_len;
3493 static inline void skb_metadata_clear(struct sk_buff *skb)
3495 skb_metadata_set(skb, 0);
3498 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3500 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3502 void skb_clone_tx_timestamp(struct sk_buff *skb);
3503 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3505 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3507 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3511 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3516 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3519 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3521 * PHY drivers may accept clones of transmitted packets for
3522 * timestamping via their phy_driver.txtstamp method. These drivers
3523 * must call this function to return the skb back to the stack with a
3526 * @skb: clone of the the original outgoing packet
3527 * @hwtstamps: hardware time stamps
3530 void skb_complete_tx_timestamp(struct sk_buff *skb,
3531 struct skb_shared_hwtstamps *hwtstamps);
3533 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3534 struct skb_shared_hwtstamps *hwtstamps,
3535 struct sock *sk, int tstype);
3538 * skb_tstamp_tx - queue clone of skb with send time stamps
3539 * @orig_skb: the original outgoing packet
3540 * @hwtstamps: hardware time stamps, may be NULL if not available
3542 * If the skb has a socket associated, then this function clones the
3543 * skb (thus sharing the actual data and optional structures), stores
3544 * the optional hardware time stamping information (if non NULL) or
3545 * generates a software time stamp (otherwise), then queues the clone
3546 * to the error queue of the socket. Errors are silently ignored.
3548 void skb_tstamp_tx(struct sk_buff *orig_skb,
3549 struct skb_shared_hwtstamps *hwtstamps);
3552 * skb_tx_timestamp() - Driver hook for transmit timestamping
3554 * Ethernet MAC Drivers should call this function in their hard_xmit()
3555 * function immediately before giving the sk_buff to the MAC hardware.
3557 * Specifically, one should make absolutely sure that this function is
3558 * called before TX completion of this packet can trigger. Otherwise
3559 * the packet could potentially already be freed.
3561 * @skb: A socket buffer.
3563 static inline void skb_tx_timestamp(struct sk_buff *skb)
3565 skb_clone_tx_timestamp(skb);
3566 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3567 skb_tstamp_tx(skb, NULL);
3571 * skb_complete_wifi_ack - deliver skb with wifi status
3573 * @skb: the original outgoing packet
3574 * @acked: ack status
3577 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3579 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3580 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3582 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3584 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3586 (skb->ip_summed == CHECKSUM_PARTIAL &&
3587 skb_checksum_start_offset(skb) >= 0));
3591 * skb_checksum_complete - Calculate checksum of an entire packet
3592 * @skb: packet to process
3594 * This function calculates the checksum over the entire packet plus
3595 * the value of skb->csum. The latter can be used to supply the
3596 * checksum of a pseudo header as used by TCP/UDP. It returns the
3599 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3600 * this function can be used to verify that checksum on received
3601 * packets. In that case the function should return zero if the
3602 * checksum is correct. In particular, this function will return zero
3603 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3604 * hardware has already verified the correctness of the checksum.
3606 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3608 return skb_csum_unnecessary(skb) ?
3609 0 : __skb_checksum_complete(skb);
3612 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3614 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3615 if (skb->csum_level == 0)
3616 skb->ip_summed = CHECKSUM_NONE;
3622 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3624 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3625 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3627 } else if (skb->ip_summed == CHECKSUM_NONE) {
3628 skb->ip_summed = CHECKSUM_UNNECESSARY;
3629 skb->csum_level = 0;
3633 /* Check if we need to perform checksum complete validation.
3635 * Returns true if checksum complete is needed, false otherwise
3636 * (either checksum is unnecessary or zero checksum is allowed).
3638 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3642 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3643 skb->csum_valid = 1;
3644 __skb_decr_checksum_unnecessary(skb);
3651 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3654 #define CHECKSUM_BREAK 76
3656 /* Unset checksum-complete
3658 * Unset checksum complete can be done when packet is being modified
3659 * (uncompressed for instance) and checksum-complete value is
3662 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3664 if (skb->ip_summed == CHECKSUM_COMPLETE)
3665 skb->ip_summed = CHECKSUM_NONE;
3668 /* Validate (init) checksum based on checksum complete.
3671 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3672 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3673 * checksum is stored in skb->csum for use in __skb_checksum_complete
3674 * non-zero: value of invalid checksum
3677 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3681 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3682 if (!csum_fold(csum_add(psum, skb->csum))) {
3683 skb->csum_valid = 1;
3690 if (complete || skb->len <= CHECKSUM_BREAK) {
3693 csum = __skb_checksum_complete(skb);
3694 skb->csum_valid = !csum;
3701 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3706 /* Perform checksum validate (init). Note that this is a macro since we only
3707 * want to calculate the pseudo header which is an input function if necessary.
3708 * First we try to validate without any computation (checksum unnecessary) and
3709 * then calculate based on checksum complete calling the function to compute
3713 * 0: checksum is validated or try to in skb_checksum_complete
3714 * non-zero: value of invalid checksum
3716 #define __skb_checksum_validate(skb, proto, complete, \
3717 zero_okay, check, compute_pseudo) \
3719 __sum16 __ret = 0; \
3720 skb->csum_valid = 0; \
3721 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3722 __ret = __skb_checksum_validate_complete(skb, \
3723 complete, compute_pseudo(skb, proto)); \
3727 #define skb_checksum_init(skb, proto, compute_pseudo) \
3728 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3730 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3731 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3733 #define skb_checksum_validate(skb, proto, compute_pseudo) \
3734 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3736 #define skb_checksum_validate_zero_check(skb, proto, check, \
3738 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3740 #define skb_checksum_simple_validate(skb) \
3741 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3743 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3745 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3748 static inline void __skb_checksum_convert(struct sk_buff *skb,
3749 __sum16 check, __wsum pseudo)
3751 skb->csum = ~pseudo;
3752 skb->ip_summed = CHECKSUM_COMPLETE;
3755 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3757 if (__skb_checksum_convert_check(skb)) \
3758 __skb_checksum_convert(skb, check, \
3759 compute_pseudo(skb, proto)); \
3762 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3763 u16 start, u16 offset)
3765 skb->ip_summed = CHECKSUM_PARTIAL;
3766 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3767 skb->csum_offset = offset - start;
3770 /* Update skbuf and packet to reflect the remote checksum offload operation.
3771 * When called, ptr indicates the starting point for skb->csum when
3772 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3773 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3775 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3776 int start, int offset, bool nopartial)
3781 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3785 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3786 __skb_checksum_complete(skb);
3787 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3790 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3792 /* Adjust skb->csum since we changed the packet */
3793 skb->csum = csum_add(skb->csum, delta);
3796 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3798 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3799 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3805 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3806 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3807 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3809 if (nfct && atomic_dec_and_test(&nfct->use))
3810 nf_conntrack_destroy(nfct);
3812 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3815 atomic_inc(&nfct->use);
3818 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3819 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3821 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3824 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3827 refcount_inc(&nf_bridge->use);
3829 #endif /* CONFIG_BRIDGE_NETFILTER */
3830 static inline void nf_reset(struct sk_buff *skb)
3832 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3833 nf_conntrack_put(skb_nfct(skb));
3836 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3837 nf_bridge_put(skb->nf_bridge);
3838 skb->nf_bridge = NULL;
3842 static inline void nf_reset_trace(struct sk_buff *skb)
3844 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3849 static inline void ipvs_reset(struct sk_buff *skb)
3851 #if IS_ENABLED(CONFIG_IP_VS)
3852 skb->ipvs_property = 0;
3856 /* Note: This doesn't put any conntrack and bridge info in dst. */
3857 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3860 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3861 dst->_nfct = src->_nfct;
3862 nf_conntrack_get(skb_nfct(src));
3864 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3865 dst->nf_bridge = src->nf_bridge;
3866 nf_bridge_get(src->nf_bridge);
3868 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3870 dst->nf_trace = src->nf_trace;
3874 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3876 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3877 nf_conntrack_put(skb_nfct(dst));
3879 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3880 nf_bridge_put(dst->nf_bridge);
3882 __nf_copy(dst, src, true);
3885 #ifdef CONFIG_NETWORK_SECMARK
3886 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3888 to->secmark = from->secmark;
3891 static inline void skb_init_secmark(struct sk_buff *skb)
3896 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3899 static inline void skb_init_secmark(struct sk_buff *skb)
3903 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3905 return !skb->destructor &&
3906 #if IS_ENABLED(CONFIG_XFRM)
3910 !skb->_skb_refdst &&
3911 !skb_has_frag_list(skb);
3914 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3916 skb->queue_mapping = queue_mapping;
3919 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3921 return skb->queue_mapping;
3924 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3926 to->queue_mapping = from->queue_mapping;
3929 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3931 skb->queue_mapping = rx_queue + 1;
3934 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3936 return skb->queue_mapping - 1;
3939 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3941 return skb->queue_mapping != 0;
3944 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3946 skb->dst_pending_confirm = val;
3949 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3951 return skb->dst_pending_confirm != 0;
3954 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3963 /* Keeps track of mac header offset relative to skb->head.
3964 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3965 * For non-tunnel skb it points to skb_mac_header() and for
3966 * tunnel skb it points to outer mac header.
3967 * Keeps track of level of encapsulation of network headers.
3978 #define SKB_SGO_CB_OFFSET 32
3979 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3981 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3983 return (skb_mac_header(inner_skb) - inner_skb->head) -
3984 SKB_GSO_CB(inner_skb)->mac_offset;
3987 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3989 int new_headroom, headroom;
3992 headroom = skb_headroom(skb);
3993 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3997 new_headroom = skb_headroom(skb);
3998 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4002 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4004 /* Do not update partial checksums if remote checksum is enabled. */
4005 if (skb->remcsum_offload)
4008 SKB_GSO_CB(skb)->csum = res;
4009 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4012 /* Compute the checksum for a gso segment. First compute the checksum value
4013 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4014 * then add in skb->csum (checksum from csum_start to end of packet).
4015 * skb->csum and csum_start are then updated to reflect the checksum of the
4016 * resultant packet starting from the transport header-- the resultant checksum
4017 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4020 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4022 unsigned char *csum_start = skb_transport_header(skb);
4023 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4024 __wsum partial = SKB_GSO_CB(skb)->csum;
4026 SKB_GSO_CB(skb)->csum = res;
4027 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4029 return csum_fold(csum_partial(csum_start, plen, partial));
4032 static inline bool skb_is_gso(const struct sk_buff *skb)
4034 return skb_shinfo(skb)->gso_size;
4037 /* Note: Should be called only if skb_is_gso(skb) is true */
4038 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4040 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4043 static inline void skb_gso_reset(struct sk_buff *skb)
4045 skb_shinfo(skb)->gso_size = 0;
4046 skb_shinfo(skb)->gso_segs = 0;
4047 skb_shinfo(skb)->gso_type = 0;
4050 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4052 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4054 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4055 * wanted then gso_type will be set. */
4056 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4058 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4059 unlikely(shinfo->gso_type == 0)) {
4060 __skb_warn_lro_forwarding(skb);
4066 static inline void skb_forward_csum(struct sk_buff *skb)
4068 /* Unfortunately we don't support this one. Any brave souls? */
4069 if (skb->ip_summed == CHECKSUM_COMPLETE)
4070 skb->ip_summed = CHECKSUM_NONE;
4074 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4075 * @skb: skb to check
4077 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4078 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4079 * use this helper, to document places where we make this assertion.
4081 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4084 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4088 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4090 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4091 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4092 unsigned int transport_len,
4093 __sum16(*skb_chkf)(struct sk_buff *skb));
4096 * skb_head_is_locked - Determine if the skb->head is locked down
4097 * @skb: skb to check
4099 * The head on skbs build around a head frag can be removed if they are
4100 * not cloned. This function returns true if the skb head is locked down
4101 * due to either being allocated via kmalloc, or by being a clone with
4102 * multiple references to the head.
4104 static inline bool skb_head_is_locked(const struct sk_buff *skb)
4106 return !skb->head_frag || skb_cloned(skb);
4109 /* Local Checksum Offload.
4110 * Compute outer checksum based on the assumption that the
4111 * inner checksum will be offloaded later.
4112 * See Documentation/networking/checksum-offloads.txt for
4113 * explanation of how this works.
4114 * Fill in outer checksum adjustment (e.g. with sum of outer
4115 * pseudo-header) before calling.
4116 * Also ensure that inner checksum is in linear data area.
4118 static inline __wsum lco_csum(struct sk_buff *skb)
4120 unsigned char *csum_start = skb_checksum_start(skb);
4121 unsigned char *l4_hdr = skb_transport_header(skb);
4124 /* Start with complement of inner checksum adjustment */
4125 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4128 /* Add in checksum of our headers (incl. outer checksum
4129 * adjustment filled in by caller) and return result.
4131 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4134 #endif /* __KERNEL__ */
4135 #endif /* _LINUX_SKBUFF_H */